Research On Robot Online Obstacle Avoidance Method In Dynamic Environment

With the development of robot technology,the collaboration between human and robot in dynamic scenarios such as service and industrial domain becomes more and more close.It is crucial to ensure that robot can generate reactive motion and avoid collision with people or other objects in the human-centered environment.Classical robot control approaches have provided us with tools to perform high speed and high precision tasks with industrial robots.However,these approaches rely heavily on hardcoded motions generated through offline planning with a limited ability to change flexibly and safely in unstructured environment.Therefore,in order to improve the adaptability and safety of robot motion in dynamic environment,this thesis studies the robot online obstacle avoidance method from three aspects: environment perception,obstacle avoidance strategy and robot control.In order to improve the adaptability of robot motion in dynamic environment,this thesis models robot reaching movements though nonlinear dynamical systems.A robust robot control strategy is established from a set of user demonstrations.In order to improve the safety of the robot motion in the dynamic environment,two obstacle avoidance strategies based on the dynamic modulation matrix and the repulsive velocity are proposed separately in this thesis.Finally,the environment perception approach based on depth sensor is proposed to extend the application scope of robot obstacle avoidance method.All of the above researches are verified by experiments on the platform of simulation or real robot.The main research contents and novelties are as follows:1.The establishment of robust robot control strategy.Gaussian Mixture Model(GMM)is used to model robot reaching movements from a set of user demonstration.In order to ensure the stability and adaptability of robot motion in the dynamic environment,Gaussian Mixture Regression is formalized as a nonlinear dynamical system,and Lyapunov function is constructed to analyze the sufficient conditions for the stability of the nonlinear dynamic system.Finally the parameter estimation problem of GMM is transformed into a constrained optimization problem to solve the local optimal model parameters.The simulation experiment on UR5 robot shows that the control strategy is robust to perturbations and adaptable to changes in dynamic environments.2.Obstacle avoidance method based on dynamic modulation matrix.The analytical expression of small obstacles is established.The dynamic modulation matrix is constructed to modify the robot motion trajectory generated by the dynamical system in real time.The velocity modulation module is designed for obstacle avoidance with robot end-effector and robot links respectively.Then this module is embedded in the robot motion generation framework based on dynamical system.The simulation results show that the proposed robot obstacle avoidance method based on the dynamic modulation matrix can achieve on-line obstacle avoidance of robot end-effector and robot links for small obstacles.3.Obstacle avoidance method based on repulsive velocity.In order to solve the problem of multi-robot safe coexistence,an obstacle avoidance method based on repulsive velocity is proposed.The model of robot platform is simplified to compute the shortest distance between robot and environment.Then the distance is used to generate repulsive velocity to achieve the effect of robot obstacle avoidance with surroundings.Experiments on a dual-arm robot platform show that the method can perform repulsive behavior of robot end-effector and robot links with environment.4.Distance estimation method based on depth sensor.In order to solve the general environment perception problem of robot obstacle avoidance in uncertain scenarios,a perception method based on depth sensor is proposed,which estimates the distance between robot and environment directly from the depth image without the need for analytical modeling of obstacles.The repulsive velocity is then generated by using the distance vector.Experiments are carried out on the UR5 robot platform.The results show that the method can perform repulsive behavior for robot links so as to avoid collision with environment.